iromu/Gemma3-1B-tools
iromu/Gemma3-1B-tools is a 1 billion parameter Gemma 3 model, fine-tuned by iromu using LoRA for enhanced tool calling and agent-style interactions. This model excels at structured function calling and multi-step agent workflows, making it suitable for small-footprint on-device or edge deployments. It significantly improves tool-calling accuracy over its base model, achieving 66% exact-args match compared to 2% for the base.
Loading preview...
Overview
iromu/Gemma3-1B-tools is a 1 billion parameter Gemma 3 model, specifically fine-tuned for tool calling and agent-style interactions. It leverages LoRA (Low-Rank Adaptation) with a google/gemma-3-1b-it base model to achieve specialized capabilities in structured function calling.
Key Capabilities
- Enhanced Tool Calling: Achieves a 66% exact-args match on validation, a substantial improvement over the base model's 2%.
- Agent-Style Interactions: Designed to facilitate multi-step conversational agents.
- Optimized for Deployment: Suitable for small-footprint, on-device, or edge deployments due to its compact size.
- Custom Chat Template: Utilizes a specific chat template for rendering tool schemas and emitting tool calls in a structured JSON format.
Training Details
The model was trained using NVIDIA NeMo AutoModel with LoRA/PEFT, employing a max sequence length of 4096 and a global batch size of 64. The training dataset was the sft_tools split of r0b0tlab/qwen3.8-max-glm5.2-kimi-k3-distillation.
Good For
- Structured tool/function calling applications.
- Developing agent-style multi-step interaction systems.
- Deployments requiring a small model footprint, such as on-device or edge computing scenarios.